nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity
9,131 Commits 1 Branch 0 Tags
634275fbbb0a01b95e86e30c5e740e7f44d10d2f
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Johannes GäßlerandGitHub a8fd165fec CUDA: lower-case PCI bus id, standardize for ggml (#22820) 2026-05-08 10:09:38 +02:00
Johannes GäßlerandGitHub 9dbb372610 Github: update issue templates (#22594) 2026-05-02 07:56:13 +02:00
Johannes GäßlerandGitHub e82aaf2587 CUDA: fix tile FA kernel on Pascal (#22541) 2026-04-30 13:04:50 +02:00
Johannes GäßlerandGitHub 739393beeb TP: fix delayed AllReduce + zero-sized slices (#22489) 2026-04-29 08:55:07 +02:00
Johannes GäßlerandGitHub 7ec36aa861 Github: set meta backend code owner (#22388) 2026-04-26 13:34:13 +02:00
Johannes GäßlerandGitHub 9725a313be CUDA: reduce MMQ stream-k overhead (#22298) 
* CUDA: reduce MMQ stream-k overhead

* use 32 bit integers for kbc
 2026-04-25 14:15:03 +02:00
Johannes GäßlerandGitHub fb19f94c71 TP: fix 0-sized tensor slices, AllReduce fallback (#21808) 
* TP: fix 0-sized tensor slices, AllReduce fallback

* fix layer structure <-> GPU count aliasing

* add missing std::fill

* fix CUDA device set, max ggml ctx size
 2026-04-20 18:09:39 +02:00
Johannes GäßlerandGitHub 4eac5b4509 CUDA: refactor mma data loading for AMD (#22051) 
* CUDA: refactor mma data loading for AMD

* fix CDNA MMQ occupancy

* fix CDNA3 mma

* fix RDNA3 compile
 2026-04-19 18:26:59 +02:00
Johannes GäßlerandGitHub fd1c0ec3f0 llama: fit ctx size for CPU only (#21568) 2026-04-18 08:16:04 +02:00
Johannes GäßlerandGitHub a6206958d2 CUDA: require explicit opt-in for P2P access (#21910) 2026-04-15 16:01:46 +02:00
Johannes GäßlerandGitHub 014dca49d6 CUDA: manage NCCL communicators in context (#21891) 
* CUDA: manage NCCL communicators in context

* add check that all backends are CUDA

* remove unused vector, limit init to > 1 GPUs

* fix warnings

* fix cuda device, cache allreduce
 2026-04-15 15:58:40 +02:00
Johannes GäßlerandGitHub ff5ef82786 CUDA: skip compilation of superfluous FA kernels (#21768) 2026-04-11 18:52:11 +02:00
Johannes GäßlerandGitHub 865ff06b2f TP: fix Qwen 3 Next data split (#21732) 2026-04-11 09:23:42 +02:00
Johannes GäßlerandGitHub 0893f50f2d common: mark --split-mode tensor as experimental (#21684) 2026-04-10 12:27:27 +02:00
Johannes GäßlerGitHubCarl Philipp KlemmGaurav GargGeorgi Gerganov
 		d6f3030047 ggml: backend-agnostic tensor parallelism (experimental) (#19378) 
* ggml: backend-agnostic tensor parallelism

* support for GPT-OSS, Qwen 3 MoE

* partial Vulkan fix

* add support for 4/8 GPUs

* unconditional peer access

* re-use buffers + ggml contexts

* fix output pattern

* NCCL support

* GGML: HIP: add RCCL support

* Remove shfl and AllReduce from backend interface

* move allocation workaround out of ggml-alloc.c

* 2d tensor set/get support

* Fix the seg fault without NCCL

* Apply suggestion from JohannesGaessler

* support for tensor dims % n_devs != 0

* fix view_offs scaling

* arbitrary num. of GPUs/tensor split

* fix compilation

* better granularity estimate

* Support device-specific host buffer types if all underlying backends expose the same type. This allows using pinned memory instead of pageable memory for CUDA.

Fix compilation errors.

* partial Qwen 3 Next support

* Fix qwen3 30b (#8)

* Fix crash with Qwen-30B-A3B Q4_0

Qwen-30B-A3B Q4_0 has an intermediate dimension of 768. Using a granularity of 256 forces an uneven split between GPUs, which is not supported by the current implementation.

* Decide block size based on tensor quantization type

* Fix crashes due to KV cache serialization (#9)

KV cache serialization requires non-zero offsets on the tensor. Add support in the meta backend to set/get a tensor with a non-zero offset.

* metal : fix build (#7)

* static memory allocations, fix usage count

* fix tensor granularity

* more even memory distribution

* use BF16 for allreduce

* rebase fixup

* better error message for unsupported architectures

* Fix device mismatch during scatter of allReduce. (#11)

There is a mismatch between the dst buffer device and the backend device, causing the use of sync copies

* Enable the previous allreduce implementation. It is better in both perf and stability (#12)

* delay AllReduce for Moe for less I/O

* build : clean-up compile warnings

* backend : move most of the meta backend API to ggml-backend-impl.h

* cont : hide unused public API in the implementation

* llama : use llama_device + remove ggml_backend_dev_is_meta()

* ggml-backend : remove unused alloc include

* minor : remove regex include

* ggml : introduce ggml-ext.h for staging new APIs

* rebase fixup

* fix tests

* llama : more robust logic for determining Meta devices (#16)

* llama : more robust logic for determining Meta devices

* cont : fix devs size check

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

* cont : fix log type

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

* disable roundtrip for meta backend

* fix arch selection

* Qwen 3.5 support

* fix Gemma 4 MoE

* fix OpenVino, SYCL

* fix test-llama-archs for CPU-only builds

* Fix Qwen 3.5 MoE

* disable meta backend tests for WebGPU

* tests : filter CPU-based devices from the Meta backend tests (#17)

* meta : formatting, naming, indentation (#18)

* formatting : llama-model.cpp

* formatting : ggml-ext.h

* formatting : ggml-backend-meta.cpp

* meta : add TODO

* add documentation

* better error messages

* fix GPT-OSS

---------

Co-authored-by: Carl Philipp Klemm <carl@uvos.xyz>
Co-authored-by: Gaurav Garg <gaugarg@nvidia.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
 2026-04-09 16:42:19 +02:00
Johannes GäßlerandGitHub a8ec0df461 llama: remove per-arch tensor name lists (#21531) 2026-04-07 15:02:03 +02:00
Johannes GäßlerandGitHub 86221cf6da CUDA: fix FA kernel selection logic (#21271) 2026-04-01 22:28:19 +03:00
Johannes GäßlerGitHubSiddhesh2377
 		36dafba5c4 llama: fix llama-model-saver (#20503) 
* llama : add fd-based model loading via llama_model_load_from_fd

* llama : address review feedback for fd-based model loading

* llama : use FILE pointer instead of fd in public API

* llama : use FILE pointer consistently, address review feedback

* fixup

* fix tensor names

* fix llama-model-saver

* roundtrip tests

* fixup

* refactor tests

* fix prints

* fix model saving

* fix CI, disable Chameleon

* print seed

---------

Co-authored-by: Siddhesh2377 <siddheshsonar2377@gmail.com>
 2026-03-25 12:53:16 +02:00
Johannes GäßlerandGitHub bd3f1d9d65 CUDA: fix BF16 FA compilation (#20865) 2026-03-22 17:53:33 +01:00
Johannes GäßlerandGitHub ae40cd27c8 CUDA: limit number of FA stream-k CUDA blocks (#20586) 2026-03-15 18:30:47 +01:00
Johannes GäßlerandGitHub a976ff081b llama: end-to-end tests (#19802) 
* tests: add end-to-end tests per model architecture

* fixup for rebase

* fix use-after-free in llama-model-loader.cpp

* fix CI

* fix WebGPU

* fix CI

* disable CI for macOS-latest-cmake-arm64

* use expert_weights_scale only if != 0.0f

* comments
 2026-03-08 12:30:21 +01:00
Johannes GäßlerandGitHub 2850bc6a13 ggml-cpu: fix data race for debug asserts (#20148) 2026-03-06 09:12:49 +01:00
Johannes GäßlerandGitHub 7f5ee54968 ggml: fix ggml_is_contiguous_n for ne == 1 (#20092) 2026-03-04 12:04:31 +01:00
Johannes GäßlerandGitHub c78e682245 CUDA: fix kernel selection logic for tile FA (#19686) 
* CUDA: fix kernel selection logic for tile FA

* add comment
 2026-02-19 12:42:58 +01:00
Johannes GäßlerandGitHub ada90bf2ba docs: ban AI for issues and discussions [no CI] (#19512) 2026-02-11 12:49:40 +01:00
Johannes GäßlerandGitHub 59377a6c87 ggml-backend: fix async set/get fallback sync (#19179) 2026-02-02 10:00:05 +01:00
Johannes GäßlerandGitHub a5bb8ba4c5 CUDA: tune GLM 4.7 Flash FA kernel selection logic (#19097) 2026-01-27 14:28:56 +01:00
Johannes GäßlerandGitHub b0311c16d2 CUDA: fix padding of GQA to power of 2 in FA (#19115) 2026-01-26 23:24:58 +01:00
Johannes GäßlerandGitHub 0c21677e43 CUDA: faster FA for GQA > 1 but not power of 2 (#19092) 2026-01-25 21:19:47 +01:00
Johannes GäßlerandGitHub e9fd8dcab4 llama-fit-params: keep explicit --ctx-size 0 (#19070) 2026-01-24 22:13:08 +01:00
Johannes GäßlerandGitHub 4e5b83b226 GGUF: check that tensor size is representable (#19072) 2026-01-24 21:57:51 +01:00
Johannes GäßlerandGitHub 8f91ca54ec CUDA: re-use MLA K data for V in MMA FA (#19057) 2026-01-24 10:09:36 +01:00
Johannes GäßlerandGitHub e2baf02162 CUDA: fix alignment check for FA (#19023) 2026-01-22 20:39:25 +01:00
Johannes GäßlerandGitHub 5c662d21a3 CUDA: fix allignment on register spill for FA (#18815) 2026-01-15 15:14:50 +01:00
Johannes GäßlerandGitHub c1e79e610f doc: ban AI-generated PR descriptions [no ci] (#18765) 2026-01-13 13:43:12 +01:00
Johannes GäßlerandGitHub d2ff4e23ac HIP: adjust RDNA3.5 MMQ kernel selction logic (#18666) 2026-01-10 17:19:01 +01:00
Johannes GäßlerandGitHub 64848deb18 llama-fit-params: free memory target per device (#18679) 2026-01-08 10:07:58 +01:00
Johannes GäßlerandGitHub 68b4d516c3 llama-params-fit: fix last devices with low VRAM (#18494) 2026-01-06 20:02:30 +01:00
Johannes GäßlerandGitHub df17a4c94f CUDA: fix FA FP16 accumulator overflow for Granite (#18614) 2026-01-05 19:51:13 +01:00
Johannes GäßlerandGitHub 0f2e42ca1d CUDA: only allocate FA tmp buffer if needed (#18564) 2026-01-03 13:55:53 +01:00
Johannes GäßlerandGitHub ecc343de63 CUDA: fix KQ max calculation (#18487) 2025-12-31 09:37:00 +01:00
Johannes GäßlerandGitHub 0bd1212a43 CUDA: fix replacment of bad archs in CMake (#18457) 2025-12-29 17:58:20 +01:00
Johannes GäßlerandGitHub e70e640db3 CUDA: Blackwell features for non-native builds (#18436) 2025-12-29 09:35:42 +01:00
Johannes GäßlerandGitHub f8d561eb87 llama-fit-params: fix step size for last device (#18415) 2025-12-28 10:52:09 +01:00
Johannes GäßlerGitHubSigbjørn Skjæret
 		e59efe6a78 github: update issue templates [no ci] (#18410) 
* github: update issue templates [no ci]

* Apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
 2025-12-28 10:50:56 +01:00
Johannes GäßlerandGitHub a4bf35889e llama-fit-params: fix overflow check (#18354) 2025-12-27 20:20:45 +01:00
Johannes GäßlerandGitHub 026d2ad472 llama: fix magic number of 999 for GPU layers (#18266) 
* llama: fix magic number of 999 for GPU layers

* use strings for -ngl, -ngld

* enacapsulate n_gpu_layers, split_mode
 2025-12-27 20:18:35 +01:00
Johannes GäßlerandGitHub a52dc60ba3 llama_fit_params: return enum for fail vs. error (#18374) 2025-12-27 09:59:19 +01:00
Johannes GäßlerandGitHub 9045c9afe5 llama-fit-params: fix Gemma 3 calculation (#18372) 2025-12-27 09:56:04 +01:00
Johannes GäßlerandGitHub 147a521636 tool/ex/tests: consistently free ctx, then model (#18168) 2025-12-22 11:00:37 +01:00
Johannes GäßlerandGitHub 0e1ccf15c7 llama: fix RPC for -fit on (#18233) 2025-12-21 19:33:08 +01:00
Johannes GäßlerandGitHub 57c1e05643 llama: offload output layer to GPU first (#18148) 2025-12-18 08:12:18 +01:00
Johannes GäßlerandGitHub 8dcc3662a2 llama-fit-params: fix memory print (#18136) 2025-12-17 21:10:03 +01:00
Johannes GäßlerandGitHub a2c199e479 common: clarify instructions for bug reports (#18134) 2025-12-17 18:44:13 +01:00
Johannes GäßlerandGitHub 6f1f6a961a Github: ask for -v logs for params_fit [no ci] (#18128) 2025-12-17 13:46:48 +01:00
Johannes GäßlerandGitHub d0794e89d9 llama-fit-params: force disable mlock (#18103) 2025-12-17 00:50:12 +01:00
Johannes GäßlerandGitHub 9dcac6cf9f llama-fit-params: lower ctx size for multi GPU (#18101) 2025-12-17 00:49:34 +01:00
Johannes GäßlerandGitHub 0e49a7b8b4 llama-fit-params: fix underflow for dense models (#18095) 2025-12-17 00:47:37 +01:00
Johannes GäßlerandGitHub 4164596c76 llama-fit-params: QoL impr. for prints/errors (#18089) 2025-12-17 00:03:19 +01:00
Johannes GäßlerandGitHub ec98e20021 llama: fix early stop in params_fit if ctx is set (#18070) 2025-12-16 14:24:00 +01:00
Johannes GäßlerandGitHub b1f3a6e5db llama: automatically set parameters not set by the user in such a way that maximizes GPU utilization (#16653) 
* llama: automatically fit args to free memory

llama-fit-params tool

* fix CI

* hints for bug reports, ensure no reallocation

* fix segfault with Vulkan

* add llama-fit-params to CI

* fix CI

* fix CI

* fix CI

* minor adjustments

* fix assignment of 1 dense layer

* fix logger not being reset on model load failure

* remove --n-gpu-layer hint on model load failure

* fix llama-fit-params verbosity

* fix edge case

* fix typo [no ci]
 2025-12-15 09:24:59 +01:00
Johannes GäßlerandGitHub 482211438d CUDA: fix overflow in MMA kernel without stream-k (#17939) 2025-12-12 17:43:58 +01:00
Johannes GäßlerandGitHub 17f7f4baad CUDA: fix unpadded strides in MMA FA kernel (#17891) 2025-12-10 12:39:56 +01:00
Johannes GäßlerandGitHub 48f47565a7 docs: clarify that CPU support should be first (#17886) 2025-12-09 20:10:36 +01:00
Johannes GäßlerandGitHub 0cdce38a97 CUDA: fix FP16 overflow in tile FA kernel (#17875) 2025-12-09 09:34:02 +01:00
Johannes GäßlerandGitHub f334b79494 HIP: fix RDNA3 FP16/BF16 matrix multiplication (#17817) 2025-12-06 13:45:36 +01:00
Johannes GäßlerandGitHub 6016d0bd41 HIP : fix RDNA4 build (#17792) 2025-12-05 13:47:52 +01:00
Johannes GäßlerandGitHub e95d0bc8fd CUDA: fix FA VKQ accumulator overflow (#17746) 2025-12-05 09:18:10 +01:00
Johannes GäßlerGitHubAman Gupta <aman>
 		2e1c9cd814 CUDA: generalized (mma) FA, add Volta support (#17505) 
* CUDA: generalized (mma) FA, add Volta support

* use struct for MMA FA kernel config

---------

Co-authored-by: Aman Gupta <aman>
 2025-12-03 16:57:05 +01:00
Johannes GäßlerandGitHub 73955f7d2a CUDA: no FP16 arithmetic for vector FA kernel (#17558) 2025-11-28 10:29:09 +01:00
Johannes GäßlerandGitHub 5d6838b74f CUDA: static assert to prevent misuse of memcpy_1 (#17198) 2025-11-12 23:13:55 +01:00
Johannes GäßlerandGitHub e14e842e87 CUDA: fix MMQ stream-k fixup ne1 indices (#17089) 2025-11-08 08:26:18 +01:00
Johannes GäßlerGitHubAman Gupta
 		6515610506 CUDA: fix should_use_mmvf for ne11 == 1 (#17085) 
* CUDA: fix should_use_mmvf for ne11 == 1

* Apply suggestion from @am17an

Co-authored-by: Aman Gupta <amangupta052@gmail.com>

---------

Co-authored-by: Aman Gupta <amangupta052@gmail.com>
 2025-11-07 20:53:14 +01:00
Johannes GäßlerandGitHub aa374175c3 CUDA: fix crash on uneven context without FA (#16988) 2025-11-06 14:05:47 +01:00
Johannes GäßlerandGitHub 22c8c3c6ad docs: explain CUDA 11 compilation [no ci] (#16824) 2025-11-06 08:14:35 +01:00
Johannes GäßlerGitHubAman Gupta
 		31c511a968 CUDA: Volta tensor core support for MMF (#16843) 
* CUDA: Volta tensor core support for MMF

* more generic checks for hardware support

* Update ggml/src/ggml-cuda/mmf.cuh

Co-authored-by: Aman Gupta <amangupta052@gmail.com>

---------

Co-authored-by: Aman Gupta <amangupta052@gmail.com>
 2025-10-31 15:57:19 +01:00
Johannes GäßlerandGitHub 7a0e900e36 llama: consistent ctx <-> buf order for KV cache (#16746) 2025-10-28 11:23:54 +01:00
Johannes GäßlerandGitHub 80d28f104c HIP: fix AMDGPU_TARGETS, update documentation (#16803) 2025-10-27 21:39:49 +01:00
Johannes GäßlerandGitHub 945501f5ea llama: fix leaked buffers for mmap + split files (#16765) 2025-10-27 09:17:31 +01:00
Johannes GäßlerandGitHub 0bf47a1dbb server: add memory breakdown print (#16740) 2025-10-23 21:30:17 +02:00
Johannes GäßlerandGitHub 51d1a8c997 CUDA: better error for FA kernel with 0 occupancy (#16643) 2025-10-21 15:27:53 +02:00
Johannes GäßlerandGitHub ee09828cb0 HIP: fix GPU_TARGETS (#16642) 2025-10-18 14:47:32 +02:00
Johannes GäßlerandGitHub 66b0dbcb2d llama-model: fix insonsistent ctxs <-> bufs order (#16581) 2025-10-17 17:41:09 +02:00
Johannes GäßlerandGitHub 9c7185dd28 CUDA: enable FA for FP32 KV cache (#16546) 2025-10-14 14:22:47 +02:00
Johannes GäßlerandGitHub 7049736b2d CUDA: fix numerical issues in tile FA kernel (#16540) 2025-10-13 17:29:45 +03:00
Johannes GäßlerandGitHub 11f0af5504 CUDA: faster tile FA, add oob checks, more HSs (#16492) 2025-10-11 20:54:32 +02:00
Johannes GäßlerandGitHub 75a3a6c2cd CUDA: refactor and deduplicate vector FA kernels (#16208) 
* CUDA: refactor and deduplicate vector FA kernels
 2025-09-27 18:45:07 +02:00
Johannes GäßlerandGitHub 4cdd0bb453 docs: fix typo [no ci] (#16244) 2025-09-25 12:12:27 +03:00
Johannes GäßlerandGitHub e789095502 llama: print memory breakdown on exit (#15860) 
* llama: print memory breakdown on exit
 2025-09-24 16:53:48 +02:00
Johannes GäßlerandGitHub 368560a1e3 CUDA: fix compilation on CC 6.0 (#16091) 2025-09-18 19:28:32 +02:00
Johannes GäßlerandGitHub c959b676be CUDA: fix FA occupancy, optimize tile kernel (#15982) 2025-09-17 15:32:42 +02:00
Johannes GäßlerandGitHub 0e6ff0046f CUDA: larger SRAM reads for tile FA, AMD FP16 dot (#15927) 
* CUDA: larger SRAM reads for tile FA, AMD FP16 dot

* fix logic for availability of v_dot2_f32_f16
 2025-09-11 21:19:58 +02:00
Johannes GäßlerandGitHub 17bc5a815f HIP: use v_dot2_f32_f16 instruction for FA (#15884) 2025-09-09 14:04:43 +02:00
Johannes GäßlerandGitHub 550cf726e1 CUDA: fix GET_ROWS for large tensors (#15882) 2025-09-09 08:11:01 +02:00
Johannes GäßlerandGitHub 79bc429262 CUDA: faster tile FA (Pascal/AMD), headsize 256 (#15769) 2025-09-07 00:26:28 +02:00
Johannes GäßlerandGitHub 01806e7771 ggml-cpu: document use of "free" memory [no ci] (#15834) 2025-09-06 13:28:44 +02:00
Johannes GäßlerandGitHub 5143fa895e CUDA: fastdiv, launch bounds for mmvq + q8_1 quant (#15802) 
* CUDA: fastdiv, launch bounds for mmvq + q8_1 quant
 2025-09-05 16:07:02 +02:00
Johannes GäßlerandGitHub c466abe158 llama: -fa 1/0/-1 aliases for -fa on/off/auto (#15746) 2025-09-02 18:17:26 +02:00
Johannes GäßlerandGitHub 5d804a4938 ggml-backend: raise GGML_MAX_SPLIT_INPUTS (#15722) 2025-09-01 16:14:55 -07:00
Johannes GäßlerandGitHub e81b8e4b7f llama: use FA + max. GPU layers by default (#15434) 
* llama: use max. GPU layers by default, auto -fa

* ggml-backend: abort instead of segfault
 2025-08-30 16:32:10 +02:00